Cofounder @ Embroidery. Building AI cybersecurity stuff.
Oslo, Norway
Joined January 2014
- Tweets44K
- Following1.9K
- Followers18.7K
- Likes64.6K
Pinned Tweet
AI "cyber models" aren't going to solve our security problems. "If we just patch all the vulnerabilities then we're safe" is a convenient story but it's also not true.
If you're worried about cybersecurity, the solution involves hard work, not just more tokens.
What did people think “focus on security” meant? It meant this. It meant finding all these problems (of which there are so many more) and fixing them.
The goal isn’t to have an easy day at the office. It’s to make sure people don’t die. It’s to stop truly horrible cyber attacks.
If you noticed that I’m just making stupid jokes today, it’s because this isn’t my area of cybersecurity. I’m not going to pretend I have something valuable to add.
There’s a lesson in there for some ai folks.
For our non-tech friends who don’t know what PHP is, it’s a hallucinogenic drug. They found a vulnerability in the drug. That’s how crazy AI is.
If OpenAI took monitoring seriously, they wouldn't build the system in this way. Why would the eval harness decide or even influence the monitoring configuration?
OpenAI isn't dumb, so it feels like they *want* this to happen...
Dario says the government should get to decide who gets access to certain capabilities, not Anthropic, and that’s why he advocates for regulation.
Except we’ve seen the bill Anthropic is proposing and it doesn’t give the government that power.
Zack Korman retweeted
If you release a patch first, it gets one PoC
If you don’t release a patch, people burn millions of tokens on you, and get many more PoCs, with new vulns you haven’t patched yet
One of these people is orders of magnitude less famous than the others.
But also funny to see how people just make stuff up about me. I’ve never been a tech skeptic.
Zack Korman retweeted
We reviewed a nonpublic version of the “Misalignment in high-stakes settings” section in Anthropic’s Aug 2026 Risk Report. We agree that risk of catastrophic harm from covered models’ misalignment is low. However, we note several significant disagreements with their analysis.
As part of our Responsible Scaling Policy, we publish regular Risk Reports. These share detailed information on the risks of our systems and how prepared we are to address them.
Our second Risk Report is now available: anthropic.com/aug-2026-risk-…
Quoted in Time explaining that if AI were truly impossible to control (it's not), then "don't worry it's a friendly aligned AI" isn't an acceptable answer.
Full article below.
Here's the full thing. It's an interesting read, trying to explain this ai safety vs security discourse. time.com/article/2026/10/20/…
There is only one answer to cybersecurity threats: Fix your shit.
Scared of threat actors using AI? Worried about rogue agents? Think abliterated models are spooky? Fix. Your. Shit.
The "ban AI models and regulate GPU access" crowd should instead fix their shit.
Zack Korman retweeted
He didn't bite, so here's Plan B just to be safe:
Here's a deep, vigorous demolition of @PalisadeAI's claims on LLMs' reproductive exhuberance.
Share etc; if you care at all for SI research to be somewhat factual it is imperative that we drive these frauds out of business.
the past two days of research led me to conclude the following:
@jeffladish/@PalisadeAI is the most blatant, obvious, disgusting liar in the blatantly, obviously, disgustingly lying world of AI misalignment research.
i'm writing this in order to bait him into asking me for proof and provide such proof in a manner visible enough to either end his career or expose his funders as either frauds or net-negative, clueless rubes.